GANDaLF: GAN for Data-Limited Fingerprinting

Abstract We introduce Generative Adversarial Networks for Data-Limited Fingerprinting (GANDaLF), a new deep-learning-based technique to perform Website (WF) on Tor traffic. In contrast most earlier work deep-learning WF, GANDaLF is intended with few training samples, and achieves this goal through the use of Network generate large set “fake” data that helps train deep neural network in distinguishing between classes actual data. evaluate low-data scenarios including as 10 instances per site, multiple settings, fingerprinting website index pages non-index within site. closed-world accuracy 87% just 20 site (and 100 sites) standard WF settings. particular, can outperform Var-CNN Triplet (TF) across all settings subpage fingerprinting. For example, outperforms TF by 29% margin 38% sets using